Rasmussen on the matter...
http://rasmussenreports.com/public_content/politics/election_20082/why_polls_sometimes_show_different_resultsWhy Polls Sometimes Show Different Results
Thursday, June 26, 2008
Since Barack Obama clinched the Democratic Presidential Nomination, most polls have shown the Illinois Senator with a modest lead over John McCain, typically around five percentage points. However, two recent polls, one by Newsweek and one by the Los Angeles Times, have shown Obama with a double digit lead.
Republicans quickly noted that the polling samples included an unusually large number of Democrats and small number of Republicans. “The case against the poll, laid out in a memo sent out today by Public Opinion Strategies, in turn sparked a response from survey director Susan Pinkus, who stood by its methodology and findings.”
At the core of this flap is a polling industry disagreement about the best way to deal with partisan identification in constructing a poll. Just about everyone agrees that party identification is one of the strongest indicators as to how a person will vote. A Republican is overwhelmingly likely to vote for a GOP candidate and a Democrat is overwhelmingly likely to do the opposite.
However, the challenge lies in finding the “right” mix of Republicans, Democrats, and unaffiliated voters. Some pollsters, including many academic and media pollsters, argue that partisan identification is fluid and changes frequently. This approach suggests that whatever partisan mix falls out from the results of a random sample is the “right” answer. In the case of the recent L.A. Times poll, this mix was 39% Democrats and 22% Republicans.
Polls that use this approach tend to produce a more volatile set of results (during Election 2004, one national firm reported results days apart that showed more than a ten-point swing in voter preference).
Others, including most political polling firms and Rasmussen Reports, argue that people rarely change their partisan affiliation (how many people do you know who consider themselves a Republican one day and a Democrat the next?). This approach produces more stable results. During Election 2004, Rasmussen Reports polling data never showed either candidate ahead by more than 3 points in our weekly data. Week-to-week changes were never bigger than a point-and-a-half.
This stable view of the electorate seems more intuitively correct. It is unlikely that large segments of voters change their views frequently during the campaign. Even today, when one-in-three voters say they could change their mind between now and Election Day, it is likely that these voters will gradually grow more or less comfortable with their default choice. It is highly unlikely that these voters will hang on every utterance by the Obama and McCain campaigns and change their preferences accordingly.
It is even more unlikely that people will change their partisan identification on a regular basis. For most Americans, party preferences are like favorite sports teams. You’re born into being a fan for one side of the other. Over time, if your team (i.e.—the Republicans) disappoints you often enough, you will drift into unaffiliated status. If things turn around (or if the other side looks even worse than you thought), you might drift back. But, the key word is drift… these decisions happen gradually over time.
Still, even if you believe that partisan identification is fairly constant, how can a pollster know what the mix of Republicans and Democrats should be?
It’s inappropriate to simply use the results from the last election. After all, a lot has changed over the past four years (and most of it has not been good for the Republican Party label). If you assumed that the mix of Republicans and Democrats this year will be the same as it was four years ago, you will end up with results far too favorable for the GOP.
At Rasmussen Reports, we address this issue by measuring changes in partisan identification on a monthly basis. We interview 15,000 people each month by telephone to dramatically reduce the level of statistical noise and get a stable result (see our latest partisan trends update and month-by-month numbers). This approach shows gradual shifts over time in keeping with the general flow of the political environment. During Election 2004, the GOP gained ground slightly as the campaign wore on. During Election 2006, the Democrats peaked at just the right time. Democrats struggled a bit (in relative terms) after taking control of Congress in 2007, but experienced a tremendous bounce during the early portions of Election 2008.
At the moment, our data shows that just over 41% of the population consider themselves to be Democrats and just under 32% are Republicans. We use the this information as the starting point for determining the mix of Republicans and Democrats in our national and state political surveys.
As a consumer of polling data, it is good to have a wide variety of information to review. It is helpful to have some polls assembled in the manner preferred by most academic and media pollsters. It is also good to have data available based upon more firmly established parameters of partisan identification. In the end, a dose of common sense and critical thinking is required.
There are currently polls showing results ranging from Obama by double digits (Newsweek and the L.A. Times) to a tie (Gallup). Most polls, including Rasmussen Reports, show something in the middle and that’s probably the best estimate of where the race stands today… and where it has been for the past few weeks.
On June 26, 2008, the Rasmussen Reports daily Presidential Tracking Poll, showed that--without leaners--Obama’s support has stayed between 45% and 47% for fifteen straight days. With leaners, he has stayed between 48% and 50% for twenty-one straight days.
As for McCain, his support has been at 40% or 41% on nineteen of the last twenty-two days. Twice, he inched up a point above that range and once he slipped a point below. With leaners, McCain’s support has stayed between 42% and 45% every day since Obama clinched the Democratic Presidential Nomination.
We don’t know where those numbers will be come November, but we expect that they will get there by moving gradually over weeks and months.
